D4: a Chinese Dialogue Dataset for Depression-Diagnosis-Oriented Chat

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 2438 - 2459 December 7-11, 2022 ©2022 Association for Computational Linguistics D4: a Chinese Dialogue Dataset for Depression-Diagnosis-Oriented Chat Binwei Yao1,2,3, Chao Shi3, Likai Zou3, Lingfeng Dai1,2,3 Mengyue Wu1,2,3∗, Lu Chen1,2,3∗, Zhen Wang4,5 and Kai Yu1,2,3 1SJTU X-LANCE Lab, Department of Computer Science and Engineering 2MoE Key Lab of Artificial Intelligence, SJTU AI Institute 3Shanghai Jiao Tong University, Shanghai, China 4Shanghai Mental Health Center 5Shanghai Jiao Tong University School of Medicine, Shanghai, China {yaobinwei, mengyuewu, chenlusz, kai.yu}@sjtu.edu.cn Abstract In a depression-diagnosis-directed clinical ses- sion, doctors initiate a conversation with ample emotional support that guides the patients to expose their symptoms based on clinical diag- nosis criteria. Such a dialogue system is distin- guished from existing single-purpose human- machine dialog systems, as it combines task- oriented and chit-chats with uniqueness in di- alogue topics and procedures. However, due to the social stigma associated with mental ill- ness, the dialogue data related to depression consultation and diagnosis are rarely disclosed. Based on clinical depression diagnostic cri- teria ICD-11 and DSM-5, we designed a 3- phase procedure to construct D4: a Chinese Dialogue Dataset for Depression-Diagnosis- Oriented Chat1, which simulates the dialogue between doctors and patients during the diag- nosis of depression, including diagnosis results and symptom summary given by professional psychiatrists for each conversation. Upon the newly-constructed dataset, four tasks mirror- ing the depression diagnosis process are estab- lished: response generation, topic prediction, dialog summary, and severity classification of depressive episode and suicide risk. Multi- scale evaluation results demonstrate that a more empathy-driven and diagnostic-accurate consul- tation dialogue system trained on our dataset can be achieved compared to rule-based bots. 1 Introduction Given the increasing worldwide health threat brought by depression, researchers have been ex- ploring effective methods for depression detection and diagnosis. Besides automatic depression de- tection from posts on social media (Orabi et al., 2018), speech (Zhang et al., 2021b) and multi- modality (Cummins et al., 2013), the dialogue thors.∗*Mengyue Wu and Lu Chen are the corresponding au- 1To get access to D4, please look at https://x- lance.github.io/D4. Figure 1: Comparison of Different Dialogue Types system is considered an effective tool for large- scale depression detection (Pacheco-Lorenzo et al., 2021). It is believed that conversation agents could reduce the concealment of sensitive information such as suicidal thoughts caused by social expec- tations (Schuetzler et al., 2018) and the emotional hindrance due to the pressure of being judged in face-to-face conversation (Hart et al., 2017). In past research, chatbots initiated for depression diagno- sis are generally implemented based on self-rating scales (Jaiswal et al., 2019; Arrabales, 2020) or diagnostic criteria (Philip et al., 2017). The final diagnosis results are obtained by asking fixed ques- tions on the scale and corresponding the user’s an- swers to each question to the scale options. These chatbots present good sensitivity and specificity in diagnosis and are more attractive and acceptable (Vaidyam et al., 2019; Abd-Alrazaq et al., 2019) than the original self-rating …

Content truncated to 500 words